Friday, October 19, 2007

Another view of the flash drive future

Startup Fusion-io was making some interesting claims when it rolled out its first flash memory card for the PCI Express bus. The card hit 100,000 IOPS which Fusion and at least one competitor said was a new high water mark. A spokesman fro Hewlett0-Packard also voiced support for the Fusion-io card in a press release from the startup.

Fusion's CTO David Flynn said the next big move is getting operating systems to work on an overdue overhaul of the old block storage I/O structures to push to a million IOPS. That will open new doors for packing the equivalent of a storage network inside an existing server blade system with the card.

I ran those claims by Michael Krause, an interconnect expert at Hewlett Packard. His initial take was that the startups performance claims could be further detailed, OSes are moving ahead in I/O, but more needs to be done. Here's a short digest of Mike's reply:

"About 8 years ago, I spoke at a conference and showed moderately high-end servers at that time doing 1 million IOPS. Without additional information on the actual performance workload, it is hard to say whether this is really impressive or not.

"Any new technology that mandates software changes, especially in a major OS subsystem, often sees significantly slower adoption than the advocates believe will occur.

"There has been significant advancements in OS I/O structures including major advances in file systems and storage subsystems. All OS support the class driver model which allows storage technology to evolve rather transparently to the OS, [and] nearly all OS also support the ability to stack or substitute file systems without impacting the rest of the OS.

"It's true that SCSI hasn't changed in years, but that has more to do with the customer demand to keep the basic block storage model intact… There is agreement among many within the industry that the traditional block storage model is limited and that perhaps moving to object based storage is the next logical progression."

Thanks, Mike.

2 comments:

Anonymous said...

This is a rather interesting prototype that needs years of maturity to become a product. Given the fact that the demo is limited to Linux indicates this is the case. Replacing Microsoft storport, passing WHQL tests and getting key customers blessing to run on mission critical servers? Good luck.

Anonymous said...

Let's not be too cynical on the chances for success - if people applied such logic to technologies such ubiquitous as Ethernet or tried to hold back the inhernet wrongness with the assertion that no one would ever need more than 640KB of memory, we'd never see progress in the industry.

There are many players entering the solid state storage market and some are quite large. Does not mean success out the door is guaranteed but it does represent a growing sense of viability. May take a few years but as they saying goes, change is in the winds, and that will eventually translate into a viable solution emerging. Here's hoping that some of these players pay attention to the basics while at a some time remain focused on progressing the industry rather than perpetuating the rather stagnant 30-year storage models and solution approach.

 
interconnects